Identification of adulterated cooked millet flour with Hyperspectral Imaging Analysis

Abstract

Hyperspectral imaging (865-1711nm) was applied to classify adulterated cooked millet flour, which contained pure cooked millet flour, pure cooked soybean flour, 25% cooked soybean flour adulterated in cooked millet flour, 15% cooked soybean flour adulterated in cooked millet flour. The spectral data were extracted from the region of interest (ROI) of samples. The effective wavelengths were obtained from spectral data by loadings of principal component analysis (PCA), successive projection algorithm (SPA) method and competitive adaptive reweighted sampling (CARS) method respectively. Least square-support vector machine (LS-SVM) was used to build classification models on full spectral data, three effective wavelengths dataset, respectively. The results showed that the overall classification accuracy of every LS-SVM model was satisfactory, over 98.5%, and classification accuracy on full spectral and effective wavelength by CARS can reach 100%. The results indicated that hyperspectral imaging could provide one accurate way for similar flour adulterated detection.